Buckwheat Detection and Classification Using Mahalanbis Distance Image and Improved Morphology Denoising Strategy

Abstract

Real-time automatic assessment of buckwheat appearance quality can improve the estimating efficiency and accuracy when buckwheat was hulled by shelling machine. In this paper, a whole kernel and nibs of buckwheat detection and classification method based on Mahalanbis distance image and improved morphology denoising strategy was proposed. This detection and classification method involves three key steps. Firstly, the whole kernel and nibs of buckwheat image was obtained by CCD camera. And Mahalanbis distance method and OTSU was used to segment the whole kernel and nibs of buckwheat. Secondly, watershed algorithm based on improved morphology denoising strategy and statistics of consecutive fields was used to separate adhesive buckwheat and count the number of buckwheat separately. Finally, color features of buckwheat was extracted and probabilistic neural network was trained to recognize the type of buckwheat. The experimental results show that the detection method based on Mahalanbis distance image and improved morphology denoising strategy can detect the whole kernel and nibs of buckwheat automatically and 98.1% of the buckwheat regions are correctly detected. The recognition rate of the whole kernel and nibs of buckwheat is higher than 92.3% when probabilistic neural network was used. The proposed method can meet the practical need of industry detection and classification.